False Data Injection Attacks Based on Least Squares Generative Adversarial Networks with Reconstruction Loss
Yichen Henry Liu, Bo Liu, Xuebo Liu, Hongyu Wu · 2024
Cybersecurity is increasingly vital in power systems, particularly with the rise of Internet of Things (IoT) devices. The integration of these devices amplifies the system’s exposure to threats like False Data Injection Attacks (FDIA). This work proposes a Generative Adversarial Networks (GANs) framework for generating FDIA against power system state estimation from the attacker’s perspective. Specifically, we propose Least Squares Generative Adversarial Networks with Reconstruction Loss (LSGAN-RL) FDIA to produce artificial measurements that pass the traditional bad data detection (BDD) mechanism. One advantage of LSGAN-RL FDIA is that it only uses the historical measurement data without knowledge of the system topology and line impedance. This algorithm adeptly mirrors the data distribution seen in power system measurements. Our simulations on the IEEE 14-bus system present a 99.31% pass rate of the crafted data evading BDD. The kernel density plot between the actual and generated measurements further underscores the effectiveness of LSGAN-RL in emulating stealthy FDIAs.